Setup, not skill
The gap between “we’ve tried AI” and “we have a working agent”
Most firms have played with ChatGPT. Very few have built anything that runs in the background, handles real data, and answers questions without someone babysitting it. That gap is not a skills gap. It is a setup gap, and this build closes it.
This is episode two of the Build AI Agents the Right Way series. By the end you will have an agent that reads your firm’s own data — held in Google Sheets or any other structured source — and answers questions through a chat interface. This episode gets the foundations in place: an n8n account, an OpenAI API key, and a live agent you can talk to within ten minutes.
The real villain
Why so many AI projects stall before they start
The blocker is not the technology. It is the advice that surrounds it. Most guidance on AI agents is written for developers, or for an audience happy to read documentation at 11pm. A Managing Partner or COO does not need to understand transformer architecture. They need something that works, fits inside their existing tools, and does not require a full-time engineer to maintain.
That is the gap n8n fills. It is an open-source automation platform with a visual, low-code interface: you connect blocks, define triggers and actions, and write no code unless you want to. For the agent we are building, we will not write a single line.
Know what you’re building
Workflows versus agents, and why the distinction matters
Before touching any tooling, be clear on what you are actually building.
- A workflow is a sequence of steps to accomplish a task. It has a trigger and a series of actions, and can be fully automated or manual. Many of the highest-value automations in a professional services firm — proposal assembly, client reporting, status updates — are pure workflows. They need good process design and the right connectors, not AI.
- An agent is different. Agents have goals and prior knowledge. They can take actions, observe the result, and adjust. They decide how to achieve a goal rather than following a pre-written script.
These are not competing ideas. A workflow can contain an agent as one of its steps; an agent can trigger a workflow. The question to ask is whether a task needs adaptive decision-making or just consistent execution. Most process automation is the latter; AI earns its place when the decision-making is the bottleneck.
AI is most valuable when the decision-making is the bottleneck.
In the KWA framework — Knowledge, Workflow, Agents — we standardise knowledge and document processes before layering in agents. Skipping that sequence is why most AI rollouts produce a promising demo and nothing else.
Ten minutes of clicks
Set up n8n and build your first workflow
n8n is free to trial — no credit card, just an email. The cloud-hosted trial runs 14 days with more than 1,000 workflow executions included, plenty for building and testing.
- Go to n8n.io, click Get Started in the top right, and follow the prompts. It will ask about your use case and company size; these help the n8n team understand their audience and have no effect on what you can build.
- Prefer to run it on your own machine? You can self-host for free using Docker, covered in a separate video.
- Once in, you land on the main workspace where you create and manage workflows.
Now build the workflow. Click New Workflow. Every workflow needs a trigger — the event that starts it. Because we are building an agent you can converse with, add the Chat Interface trigger; no configuration needed yet. Click the + node after it, search for AI Agent, and add it. Leave the prompt settings as the defaults for now — writing a professional, bespoke system prompt comes in a later episode. Finally, inside the AI Agent node, find the LLM option and select OpenAI. That is where the API key connects.
Connect the brain
Get your OpenAI API key and wire it in
The agent needs a language model to think with.
- Go to platform.openai.com. New accounts include free credits, which is enough for everything in this series.
- In the dashboard, create a new API key. The name is for your own reference; if you plan to issue multiple keys, name them by purpose or by the person they belong to. Copy the key once created — you will not be able to see it again.
- Back in n8n, paste the key into the OpenAI credential field and press Save. n8n tests the connection immediately and confirms it is working.
See it run
Test the agent, and what comes next
Your workflow now has two nodes: a Chat Interface trigger and an AI Agent connected to OpenAI. Click Test Workflow, a chat panel opens, and you can send a message. At this stage the agent has no access to your data and no custom instructions, so its answers will be generic — but it is running, it is responding, and you built it without touching a line of code.
That is the Workflow layer of the KWA framework in place. The next steps give it Knowledge to draw on and configure the Agent layer to use that knowledge purposefully:
- Writing a system prompt that gives the agent a clear role, tone, and boundaries
- Connecting the agent to a Google Sheet so it can answer questions about your own data
- Adding tools that let the agent take actions, not just answer questions
- Deploying the agent for your team with appropriate access controls
The goal is not a clever demo. Building an agent no longer requires a developer or a six-figure budget — the tooling exists, the costs are low, and the setup is measured in minutes. The 5 Steps for AI Leadership framework starts with Align for a reason: before you build anything, you need to know what you are building it for, and your team needs to understand why. An agent your senior team trusts to give reliable answers — because it was set up deliberately, tested against real briefs, and given appropriate guardrails — is a competitive advantage.